Galaxy-BioProd: artificial intelligence invests in the field of retro-biosynthesis: a state of the art

Retro-synthesis is a method that involves breaking down a molecule into its basic components to synthesize it. This approach has a long history in chemistry, and retro-biosynthesis, a variant, is also used in fields such as biocatalysis and synthetic biology. Artificial intelligence (AI) is opening up new perspectives in this field today. However, this approach is complex to implement, as it requires precisely identifying possible biological and chemical reactions, as well as the enzymes capable of catalyzing them.

Retro-Biosynthesis ?

Retro-biosynthesis, a variant of classical retro-synthesis, relies on reactions catalyzed by enzymes, often derived from biological systems such as bacteria. Unlike classical chemical synthesis, these processes are highly precise, consume less energy, and generate minimal waste.

A concrete example is the production of precursors for antibiotics or recyclable materials. These approaches reduce reliance on conventional chemical processes, which are often energy-intensive and polluting. However, current technological limitations make planning these biological pathways highly complex. The goal is to start with a target molecule, for example, a drug, and identify a series of enzymatic reactions leading to its production. AI tools provide invaluable assistance here by rapidly exploring a large number of possible chemical and biological scenarios.

AI in Retro-Biosynthesis

Identifying molecular building blocks is essential for synthesizing molecules. This is where artificial intelligence comes in as a powerful tool. With its ability to analyze vast databases and predict complex patterns, AI accelerates the discovery of innovative synthesis pathways.

Take the example of breaking down a complex molecule, such as an antibiotic. AI can identify biologically available building blocks, such as glucose or amino acids, using two approaches. The first involves leveraging existing databases to identify known reaction mechanisms. The second relies on algorithms based on so-called generative models, capable of directly predicting the necessary compounds. These approaches significantly reduce the time and costs associated with developing new synthesis strategies.

As for the synthesis of complex molecules, it often involves several steps. In this context, AI acts like a chemical GPS, navigating a vast network of possibilities to choose the optimal path. Sophisticated algorithms, such as reinforcement learning, evaluate thousands of potential combinations, considering criteria such as enzyme availability and reagent costs. This streamlined approach optimizes the synthesis process.

What does the research say ?

In their work, the authors adopted a systematic approach to compile current knowledge on retro-biosynthesis. They drew inspiration from established methodologies for selecting scientific publications. This involved exhaustive research covering numerous scientific disciplines (biology, chemistry, computer science) and integrating information from broader sources than scientific articles, such as conference proceedings. The objective: to provide a clear and as comprehensive as possible view of the current state of knowledge.

Although promising, it appears that AI in retro-biosynthesis still faces significant challenges. Available data is sometimes incomplete, and AI cannot yet accurately predict all biological variables, such as enzymatic stability under industrial conditions.However, the integration of new data and the development of more powerful algorithms seems to be a promising evolution to make these tools indispensable in the near future, for both research and industry.

What can we conclude from this?

Retro-biosynthesis, supported by AI, offers a promising future for creating more sustainable and innovative solutions. Whether in drug production, green chemistry, or the development of new materials, it represents a bridge between technological innovation and positive human impact. The path remains complex, but with a collaborative approach, current challenges can turn into opportunities for scientists and industry professionals.

To Learn More:

Guillaume Gricourt, Philippe Meyer, Thomas Duigou, Jean-Loup Faulon. Artificial Intelligence Methods and Models for Retro-Biosynthesis: A Scoping Review. ACS Synthetic Biology, 2024, 13 (8), pp.2276-2294. ⟨10.1021/acssynbio.4c00091⟩. ⟨hal-04673511⟩

Galaxy-BioProd project.